US 20050055110 A1
A process simulation system that simulates the operation of the manufacturing and measurement systems used to produce and measure the articles being analyzed relative to engineering design targets, engineering design tolerances, producibility and/or quality. In one embodiment, the user is able to assess, without risk or production cost while accelerating speed-to-market, the effect of contemplated changes (i.) to engineering design targets, (ii.) to engineering design tolerances, (iii.) to tooling, (iv.) to part pre-process dimensions and (v.) to the measurement system—on manufactured part dimensions, producibility and quality (i.) without modifying tooling, (ii.) without changing part pre-process dimensions, (iii.) without producing new parts, (iv.) without measuring article characteristics on the new parts and (v.) without changing the measurement system. The simulation functionalities, according to embodiments of the present invention, enable the user to verify whether or not the contemplated changes will have the desired effect without incurring the time and expense involved in actually making the changes, producing parts, measuring part characteristics, changing the measurement system and then determining whether the changes accomplished the desired objectives.
1. A method facilitating analysis of a process directed to the creation of a set of articles, wherein the set of articles exhibit a range of variation as to a plurality of article characteristics, the method comprising
identifying a model of the process, wherein the model characterizes the regression relationship between at least two article characteristics in the plurality of article characteristics;
receiving at least one modification parameter; and
changing the model of the process based on the at least one modification parameter.
2. The method of
3. The method of
4. The method of
5. The method of
6. The method of
7. The method of
8. The method of
displaying a graphical representation of the model relative to the at least one design specification value.
9. The method of
10. The method of
11. The method of
12. The method of
13. The method of
14. The method of
15. The method of
16. A method facilitating analysis of a process directed to the creation of a set of articles, wherein the set of articles exhibit a range of variation as to a plurality of article characteristics, the method comprising
identifying a model of the process, wherein the model includes a regression model between a first article characteristic and a second remaining article characteristic in the plurality of article characteristics, and wherein the model contains one or more instances of any one of the following: a design target corresponding to an article characteristic, a lower specification limit for an article characteristic, an upper specification limit for an article characteristic;
receiving at least one modification parameter; and
changing the model based on the at least one modification parameter.
17. The method of
18. The method of
19. The method of
20. The method of
21. The method of
22. The method of
23. The method of
24. The method of
25. The method of
26. A method facilitating analysis of a process directed to the creation of a set of articles, wherein the set of articles exhibit a range of variation as to a plurality of article characteristics, the method comprising
identifying a model of the process including a regression model(s) between a predictor characteristic selected from the plurality of article characteristics and at least one remaining article characteristic in the plurality of article characteristics,
receiving at least one modification parameter; and
changing the model of the process based on the at least one modification parameter.
27. The method of
28. The method of
receiving a design specification value for the predictor characteristic and a design specification value for said at least one of the remaining article characteristics; and wherein the design specification value for the predictor characteristic and the design specification value for said at least one of the remaining article characteristics are located relative to the regression model(s) between the predictor characteristic and said at least one remaining article characteristics.
29. The method of
displaying a graphical representation of the model.
30. The method of
31. The method of
32. The method of
33. The method of
34. The method of
35. The method of
36. The method of
37. The method of
The present application is a continuation-in-part of U.S. patent application Ser. No. 10/752,843, which is a continuation of U.S. patent application Ser. No. 10/357,690 filed Feb. 4, 2003 and entitled “Manufacturing Design and Process Analysis System,” which is a continuation-in-part application of co-pending and commonly-owned U.S. patent application Ser. No. 10/067,704 filed Feb. 4, 2002 and entitled “Manufacturing Design and Process Analysis System,” both of which are incorporated herein by reference for all purposes. The present application also claims priority from U.S. Provisional Application Ser. No. 60/491,148 filed Jul. 30, 2003 and entitled “Manufacturing Design and Process Analysis and Simulation System.”
The present invention relates to manufacturing, design, tooling, and process engineering and, in one embodiment, to methods, apparatuses and systems facilitating the design, tooling, production and/or measurement tasks associated with manufacturing and other processes. In one embodiment, the present invention relates to decision-making and logic structures, implemented in a computer software application, that facilitate all phases of the design, development, tooling, pre-production, qualification, certification, and production process of any part or other article that is produced to specification.
The world of manufacturing, including process engineering, has been under continuous and accelerating pressure to improve quality and reduce costs. This trend shows signs of further accelerating rather than decelerating. From a manufacturing perspective, quality refers to producing parts that 1.) are close to or at engineering design targets, and 2.) exhibit minimal variation. The realm of design engineering has also been under continuous pressure to improve quality and reduce costs. Design engineering must create nominal design targets and establish tolerance limits where it is possible for manufacturing to produce parts that are 1.) on target and 2.) that fall within the design tolerance limits. In-other-words, engineers are tasked not only with designing articles to meet form, fit and function, but with designing them for producibility.
In any manufacturing or other process, there are five fundamental elements (see
The traditional approach to producing articles, such as parts or other components, that meet design specifications is a logical one based on a search for causation. This approach is based on the principle that, control over the variables that influence a process yields control over the output of that process. In-other-words, if one can control the cause, then one can also control the effect.
Plastic injection molding is an example of this situation. With at least 22 control variables, even when these control settings have only two levels each (a high and a low temperature, a high and a low pressure, etc.), there are nevertheless over 4 million possible combinations. Indeed, there are billions of possible combinations when three levels (high, medium and low settings) are considered. Furthermore, changes to process variables may have varying effects on the resulting article characteristics; for example, increasing a pressure setting can increase a first article characteristic, decrease a second, and not affect a third. Simple interactions, complex interactions and non-linearities complicate the situation further. In addition, there are usually multiple mold cavities in a single mold. Finally, there are numerous article characteristics (dimensional, performance, or other requirements) that must be met. In light of the preceding, it is often extremely difficult to establish the combination of factors from the large number of part design targets, part tolerance limits, mold design characteristics and injection molding press settings that produces acceptable articles.
Some progress has been made in this regard. Design of Experiments (DOE) methodology greatly reduces the number of experiments that must be conducted to understand the impact of a selected subset of control variables on the resulting output of a process. Unfortunately, even after performing a designed experiment, there are still a large number of control variables that can affect the resulting articles. In any event, extensive measurement of produced parts is still conducted by both the supplier and the OEM customer to ensure that acceptable articles are produced.
In addition, there are two main paths to achieving improved manufacturing quality. The first is to measure the parts after they are produced and then compare the parts to specification requirements (design targets and tolerances). This is an “on-line” process utilizing feedback. The parts are usually measured, to some extent, by both the producer and the customer (OEM, first tier manufacturer, second tier manufacturer, etc.). Measuring the parts, recording and analyzing the data, and reporting the results, however, is a very expensive and resource consuming process.
In their efforts to improve quality, many manufacturers have begun to use the techniques of Statistical Process Control (SPC) and Process Capability studies. Indeed, many customers require their suppliers to perform SPC or other equivalent measurement, recording, analysis and reporting procedures. According to this technique, samples are taken from the production line, measured and then analyzed to see if any abnormal (not normally distributed) patterns or data points emerge. If such abnormal data points are detected, the process is considered “out-of-control” (i.e., failing to yield a consistent predictable output) and production is immediately stopped to fix the process. The measurement data from manufactured parts is analyzed using sophisticated SPC statistical methods and charting tools embodied in specialized computer programs. Since most parts have many different dimensions, measurement and SPC analysis have usually been applied to a large number of part dimensions for each part, increasing the time and expense associated with production. However, SPC is far less expensive in the long run than shipping unacceptable parts and/or having to sort acceptable parts from unacceptable parts.
It has also been difficult for manufacturers (and their customers) to determine 1.) what percentage of the dimensions should be monitored using SPC and 2.) which dimensions should be measured if the full set of dimensions is not monitored. Usually, most, if not all, of the “critical” dimensions called out by the design engineer are measured and analyzed using SPC techniques. However, economic constraints can result in fewer than the desired number of dimensions being measured and analyzed. Guesswork is then frequently involved as to which dimensions to select for SPC or other analysis.
A second path to improving manufacturing quality is by reducing the natural variation in the manufactured articles. The accuracy of maintaining the process control factors can be improved and/or the “noise” factors can be eliminated or minimized. This is an “off-line” process improvement using feed-forward. Reducing natural variation is also an expensive proposition since many relatively small common causes of variation are present. The degree to which the natural variation in the parts produced must be reduced is usually determined through expensive process capability studies, usually conducted on each “critical” dimension.
In light of the foregoing, a need in the art exists for methods, apparatuses and systems facilitating design and manufacturing processes and, more particularly, addressing the problems discussed above. For example, a need in the art exists for methods and systems that allow for reductions in time and cost associated with the measurement, recording, analysis and reporting processes discussed above in connection with, for example, SPC studies, Process Capability studies, shipping inspection and receiving inspection. A need in the art exists for methods to determine how to adjust inputs to a process in order to achieve the desired outputs. A need in the art also exists for methods and systems facilitating a determination of how many article characteristics (e.g., dimensions, performance measures, etc.) should be measured for a given process. Lastly, a need in the art exists for methods and systems that enable an assessment of which article characteristics should be measured for a given process. As discussed in more detail below, embodiments of the present invention substantially fulfill these needs.
The present invention provides methods, apparatuses and systems that facilitate the design, production and/or measurement tasks associated with manufacturing and other processes. In one embodiment, the present invention relates to decision-making and logic structures, implemented in a computer software application, facilitating all phases of the design, development, tooling, pre-production, qualification, certification, and production process of any part or other article that is produced to specification. In one embodiment, the present invention provides knowledge of how the multiple characteristics of a given process output are related to each other, to specification limits and to pre-process inputs. This knowledge facilitates a reduction in measurement, analysis and reporting costs both prior to and during production. In various implementations, the present invention allows for one or more of the following: It determines the changes needed to pre-process inputs in order to achieve production at design targets. It provides a prioritized order for relaxing design tolerances. It assesses the feasibility of producing parts that meet specification limits. It assesses the trade-off between performance and producibility and provides design targets that improve producibility. It provides a determination of when process variability needs reduction. It facilitates material comparison and selection. It provides process engineers and operators with improved operating guidelines.
The present invention uses analytical techniques to accomplish the preceding objectives and advantages. As discussed below, graphical techniques, in one embodiment, can optionally be used in addition to, or in place of, analytical techniques. Graphical techniques, including but not limited to charts, graphs, and plots, can also be used to display analysis results. The present invention employs powerful statistical methodologies that, in one embodiment, allow for a determination of which and how many article characteristics should be measured, potentially reducing the cost and resource expenditure associated with measurement, recording, analysis and reporting. Embodiments of the present invention also assist design engineers in designing articles for producibility. Embodiments of the present invention can also be configured to provide critical information necessary for design engineers and tooling engineers to modify design requirements for process inputs in order to make it possible for manufacturing to hit design targets and stay within specification tolerance limits. Embodiments of the present invention can also be employed to identify, using a systems engineering approach, which article characteristics have the most restrictive targets and specification tolerance limits. Such information, for instance, allows for an evaluation of whether or not tolerances should be increased and, if so, which tolerances and on which article characteristic. The present invention can also be employed to reduce the cost of performing process capability studies by reducing, in some cases dramatically so, the number of process capability studies that must be conducted. These and other aspects of the present invention will be become apparent from the following description of preferred embodiments of the present invention.
In one embodiment, the present invention provides methods, apparatuses and systems that facilitate understanding and analysis of how the relationship between predicted article characteristics and one or more predictor article characteristics reflects the ability of a process to achieve a desired objective (e.g., the production of parts at target specification and/or within specification tolerances). In one embodiment, the present invention provides a process analysis system that generates a set of graphs and/or tables (such as Constraint Tables, Offset Tables and Relaxation Tables) that enable the user to understand and analyze the relationship among article characteristics to make powerful and informed decisions as to potential design or tooling, or pre-process dimensional or process changes. As discussed in more detail below, the relationship between the predictor characteristic and one or more given predicted article characteristics can be generally categorized into one of three possible situations: 1) where it is possible to produce an article outside of specification limits as to the predicted article characteristic; 2) where the predicted article characteristic is robust and will always be inside of specification limits; and 3) where the predicted article characteristic constrains the predictor article characteristic. In one embodiment, the present invention provides a methodology for analysis of the possible relationships between the predictor characteristic and the remaining article characteristics in order to determine, for example, which predicted article characteristics can be safely ignored, which predicted article characteristics will constrain the operating range or window relative to the predictor characteristic, and which predicted article characteristics could result in producing articles outside of design specification limits. With such a categorization of predicted article characteristics, users (such as design engineers, tooling engineers, process engineers, inspectors and the like) are then in position to make decisions as to how to treat each predicted article characteristic. For example, in the case where it is possible to produce defects, the user may decide to relax specification tolerances, and/or modify pre-process inputs, and/or constrain process variables. In addition, if a given predicted article characteristic is robust (within specification) for all possible values (at least within design specification limits) of the predictor article characteristic(s), the predicted article characteristic can be ignored, for example, during post-processing tasks such as part measurement and analysis associated with pre-qualification, qualification, certification and production activities. The present invention also facilitates analysis of the impact of predicted article characteristics that constrain the allowable range for the predictor article characteristic(s). As discussed in more detail below, the present invention allows a user to consider the impact of a constraining predicted article characteristic, and degree of constraint, on the producibility of the article as called out in a given design specification, and also allows for an assessment of the suitability of a given process to generate output that meets design specifications/requirements. As discussed in more detail below, embodiments of the present invention are operative to generate graphs, tables and charts, such as Scatter Charts, Constraint Tables, Offset Tables and Relaxation Tables, that support the analytical and decision-making tasks associated with manufacturing and other processes.
In one embodiment, the present invention further provides a design and process analysis system that simulates the operation of the manufacturing system used to produce the articles being analyzed and the measurement system used to measure the article characteristics, including the manufacturing process, the measurement process, tooling, pre-process part dimensions, material responses and natural variation. In one embodiment, the user is able to assess the effect of contemplated changes—(i.) to engineering design targets, (ii.) to engineering design tolerances, (iii.) to tooling, and (iv.) to part pre-process dimensions—on manufactured part dimensions (i.) without modifying tooling, (ii.) without changing part pre-process dimensions, (iii.) without producing new parts and (iv.) without measuring article characteristics on the new parts. The simulation functionality, according to embodiments of the present invention, enables the user to verify whether or not the contemplated changes to design target, design tolerances, tooling and part pre-process dimensions will have the desired effect. This simulation and verification is done without incurring the time and expense involved in actually making the changes, producing parts, measuring part characteristics and then determining whether the changes accomplished the desired objectives. In one embodiment, the user inputs (i.) correction factors for tooling, and/or (ii.) correction factors for part pre-process dimensions, and/or (iii.) revised values for design targets, and/or (iv.) revised values for design tolerances. The process simulation and analysis system then simulates the manufacturing process operating with corrected tooling and/or corrected pre-process part dimensions, creating a new, simulated set of part dimensional data. This new set of dimensional data along with revised(if any) engineering design targets and tolerances is then analyzed using the process analysis system to verify that the desired results were obtained from the changes.
A. Principles and Concepts
The present invention utilizes several graphical, statistical and mathematical techniques directed to analyzing the relationship between article characteristics to achieve a novel design and manufacturing process analysis system. Among these are scatter diagrams, correlation coefficients, coefficients of determination, linear, non-linear and multi-variate regression, prediction intervals, adjusted prediction intervals, prediction using regression, prediction using prediction intervals, DOE, averages and weighted averages.
In understanding the difference between the prior art exemplified in
Injection molding processes typically have relatively small natural variations as compared to typical measurement error. Consequently, the first method of generating injection molded parts for evaluation of relationships can be unproductive. Consequently, the second method for generating parts is more applicable for injection molded parts. However, other processes may exhibit sufficient natural variation in order to use the above-identified method. According to a second method, variation in part characteristics is induced. In the case of injection molding, variation is induced by deliberately changing process control settings. In this manner, the variation in part characteristic values becomes large relative to measurement error. Correlations, to the extent that they exist, then become apparent.
As previously mentioned, DOE is a method that assists in reducing an unmanageably large number of experimental conditions down to a manageable few experimental conditions. Since variation must be induced in the field of injection molding, there is utility in using DOE techniques to design an efficient experiment. Use of this method has further utility in that there are commercially available computer applications that efficiently analyze the data and report the results of the analysis. Thus, one beneficial by product of using DOE, is that useful information may be extracted from an experimental run. In particular, it is usually possible to identify at least one process control setting that can be used to significantly affect the part characteristics of resulting output. The information obtained from DOE has utility as it can be used to adjust a process control setting to achieve a desired change in the joint operating position of the part characteristic values along the regression model, as explained below.
There is a second advantage to inducing variation in an experimental run that is not connected with any efficiency measure associated with using DOE. This second advantage lies in the fact that the present invention, in one embodiment, identifies the process control settings that have the greatest impact or influence on the part characteristics. The present invention may also rely, in part, on the experience of the injection molding press operators and associated manufacturing and process personnel to select those “thigh impact” control settings. It should be noted that in injection molding, the usual paradigm is to minimize changes to the press settings. In contrast, the present invention seeks to maximize their impact for purposes of inducing part variation for further analysis. In-other-words, for the purpose of inducing variation, the present invention seeks out the “worst” control settings. The “worst” control settings from a production perspective become the “best” control settings from the perspective of inducing variation.
As previously noted, there are a large number, typically 22 or more, process control settings in the field of injection molding. The present invention, in one embodiment, incorporates “scientific” or “decoupled molding” principles to identify the high impact press controls. As with DOE, it is not necessary to use “scientific”/“decoupled molding” principles, but it potentially provides additional identification advantages. Thus, when several, typically 3-5, of the highest impact control settings are changed in the experimental run, the greatest amount of variation will be introduced into the part characteristics. This variation will be of two types. The first will be a translation of the joint operating position along the regression line. The second may induce scatter of the data points about the regression line. It is important to create a robust data set to yield, in turn, a robust regression model for prediction.
Finally, the use of DOE techniques provides additional information. Specifically, the use of DOE techniques to induce part variation further allows for an understanding of how the process control variables that were changed affect part characteristics and potentially how these control variables interact with each other.
As previously discussed, it is difficult to establish the relationship between injection molding control settings and part characteristics for several reasons including the large number of control variables, the potentially large number of part characteristics, simple interactions, complex interactions, non-linearities and other effects. One of the great utilities of this invention is that even though there may be many process control variables that influence a part characteristic, and those changes may influence any one part characteristic in a very complex manner, changes in these variables have a predictable effect on the relationship between the predictor characteristic and at least one remaining article characteristic. Accordingly, as discussed in more detail below, the systems and methods of the present invention allow design engineers and process operators to rely on the values of a predictor characteristic in order to determine whether one or more predicted characteristics complies with design specifications. In addition, the systems and methods of the present invention allow design engineers and process operators to focus on the predicted characteristic in efforts to adjust process output to comply with design specifications. These and other advantages will become apparent from the description provided below.
The regression model of
Analysis of process output in this manner provides a variety of useful information facilitating design and manufacturing processes. For example,
For didactic purposes, the description of preferred embodiments primarily details application of an embodiment of the present invention to injection molding processes. The present invention, however, has application to a variety of manufacturing processes, such as plating, semiconductor manufacturing, machining, and any other process where material is added, subtracted, or otherwise changed in form or structure. In addition, the present invention can be applied to aid the design of a manufactured article, the development of a process to manufacture the article, and/or the reduction of measurement costs. Moreover, the present application has application to a variety of articles, including stand-alone articles or items, as well as articles intended as components, elements or parts of a combination. Accordingly, the description of the preferred embodiments set forth herein refers to “articles” and “parts” interchangeably.
The present invention also has application in assessing the impact of sources of variation other than variation caused by changes in press control settings. Virtually any source of variation, if it causes sufficient variation in the part characteristic value, can be assessed. Selected examples could include determining the effect of setup-to-setup variation, determining the effect of press-to-press variation for injection molding, determining temporal effects such as the impact of seasonal effects and assessing the impact of different types of raw material or the impacting of purchasing either raw material or components from different suppliers.
In addition, embodiments of the present invention can be performed without the aid of a computing device, such as a personal computer, to perform various mathematical and statistical computations set forth herein. For a small number of article characteristics, it is entirely feasible to do all of the analysis and/or graphing by hand and/or with a spreadsheet. In a preferred embodiment, however, given the large amounts of data and computational requirements, various operations associated with the present invention are performed with a computing device configured to execute the operations described herein and display resulting data on a user interface display.
B. Exemplary System Architecture
In one embodiment, the above-described system architecture operates in connection with computer hardware system 800 of
The elements of computer hardware system 800 perform their conventional functions known in the art. In particular, network/communication interface 824 is used to provide communication between system 800 and any of a wide range of conventional networks, such as an Ethernet, token ring, the Internet, etc. Mass storage 820 is used to provide permanent storage for the data and programming instructions to perform the above described functions implemented in the system controller, whereas system memory 814 is used to provide temporary storage for the data and programming instructions when executed by processor 802. I/O ports 826 are one or more serial and/or parallel communication ports used to provide communication between additional peripheral devices which may be coupled to hardware system 800.
Hardware system 800 may include a variety of system architectures and various components of hardware system 800 may be rearranged. For example, cache 804 may be on-chip with processor 802. Alternatively, cache 804 and processor 802 may be packed together as a “processor module”, with processor 802 being referred to as the “processor core”. Furthermore, certain implementations of the present invention may not require nor include all of the above components. For example, the peripheral devices shown coupled to standard I/O bus 808 may be coupled to high performance I/O bus 806; in addition, in some implementations only a single bus may exist with the components of hardware system 800 being coupled to the single bus. Furthermore, additional components may be included in system 800, such as additional processors, storage devices, or memories.
In one embodiment, the elements of the present invention are implemented as a series of software routines run by hardware system 800 of
A. Generating a Set of Articles Having a Range of Variation as to a Plurality of Article Characteristics
As described above, the present invention assesses the relationship between article characteristics associated with a set of articles having a range of variation as to the article characteristics. According to one embodiment of the present invention, a user generates a set of parts having a range of variation as to a plurality of article characteristics according to a given process. For example, a user may install an injection molding tool in an injection molding machine and produce a set of articles. The set of articles, or a sample thereof, are then measured or otherwise inspected or assessed as to the article characteristics of interest. The resulting set of data is then recorded (e.g., such as in an Excel spread sheet table) and used for subsequent analysis.
A variety of article characteristics can be measured and analyzed. For example, measured or otherwise determined article characteristics can include the dimensions of the article (e.g., length, height, width, circumference, overall diameter, etc. of the article or a given feature of the article), hardness, porosity, bowing, smoothness, void characteristics (whether voids exists and their number), color, strength, weight, and any other article characteristic, including performance characteristics such as the spray pattern of a nozzle or flow rate through a hydraulic restrictor.
As discussed above, the present invention can be applied to a set of articles where variation of the article characteristics occurs naturally or is induced by varying process control variables associated with the process that creates the article. When articles are produced with unchanged process control settings, there is typically little natural variation in the resulting article characteristics. This is particularly true for injection molded plastic parts. Measurement error may obscure or otherwise render unreliable the natural variation observed from a given set of articles. If it is not cost effective to use more precise measuring instruments, then variation should be induced in the parts by varying process settings. Accordingly, in a preferred embodiment, article variation is induced when measurement error is large compared to natural part variation.
A.1. Inducing Variation
Article variation can be induced by varying process settings. To induce variation in a preferred form, the operator varies the process settings during the manufacturing process and allows the process to come to equilibrium between setting changes before manufacturing parts for measurement. In addition, the operator in a preferred embodiment of the method selects the set or subset of process settings that induces the greatest variability in the article characteristics of interest. In a preferred embodiment, the upper and lower limits for the process settings are chosen such that the process produces parts without harming the process equipment or tooling. Moreover, in a preferred form, the magnitude of the changes in process settings is chosen to induce variation across the full range between the article characteristic upper and lower specification limits for each of the article characteristics of interest.
Selecting which process settings to vary can be accomplished using several alternative methods. In one embodiment, settings can be selected randomly. In a second embodiment, settings can be selected based on the experience of the process engineer or process operator. In the third embodiment, settings can be selected by a process expert. For example, many experts in plastic injection molding will recommend that mold temperature, hold time and hold pressure are three process settings that induce significant variation in part characteristics. In a fourth embodiment, a number of settings can be varied using a Design of Experiments (DOE) screening design. The results from this experiment can then be used to determine which process settings have a significant impact on part characteristics. The smaller number of high impact process settings can then be varied to induce variation in the part characteristics.
As to injection molding processes, part variation, in one embodiment, may also be induced by selecting and varying process control settings using scientific/decoupled molding techniques. Scientific/decoupled molding techniques provide a method of reducing the large number of press settings down to three or four key variables. Furthermore, scientific/decoupled molding techniques can be used in conjunction with the experience of the mold press operator to determine which press settings should be varied. In a preferred embodiment, the set of articles produced comprises articles from an adequate number of repetitions at each set of process control settings.
In a preferred embodiment, Design of Experiments (DOE) methodology is employed to generate a set of articles having a range of variation. DOE can be used irrespective of whether the determination of which process settings to vary is made using operator experience, decoupled/scientific molding principles, or some combination thereof. DOE defines efficient experimental setups that allow the extraction of the maximum amount of information for a relatively small experimental effort. Once it has been determined which press settings to change, DOE defines efficient experimental setups that allow the extraction of the maximum amount of information for a relatively small experimental effort. This applies to both the design of the experiment (e.g., combinations of process settings, and the number of replications at each combination, etc.) and to the analysis of the data. A wide variety of known DOE techniques and available software tools can be used to design the experimental run that induces part variation.
As discussed in more detail below, the use of DOE to produce a set of articles for analysis provides “bonus” information that can be used, after analysis according to the present invention, to move a given article output closer to target and to reduce variation in the articles. For example, such information allows the operator to adjust press settings to accomplish the following during production: 1) move product output to target, and/or 2) minimize product variation, and/or 3) minimize cost, and/or 4) minimize press cycle time.
A.2. Receiving Article Characteristic Values
In one embodiment, the present invention is implemented by a computing device (such as a special-purpose or general purpose computer) configured to execute the functionality described herein. After a given set of articles is produced and article characteristics are measured, in a preferred form, a suitably configured computing device, executing data input module 102, receives the article characteristic values associated with the set of articles and stores them in memory.
In addition, as discussed below, data input module 102 is operative to receive other data associated with operation of embodiments of the invention. For example, data input module 102 is operative to receive target values, as well as upper and lower specification limits, for all or a subset of article characteristics. In one embodiment, such data is used to provide users the ability to assess the relationship between process output and design specifications for a given set of process inputs.
B. Assessing Relationship between Article Characteristics
To allow for an assessment of the relationship between article characteristics associated with a set of articles, process analysis application 100, in one implementation, generates a set of scatter diagrams each based on a pair of article characteristics. See
In one embodiment, display module 108 generates graphical displays including scatter diagrams for presentation on display device 818 to allow the user to visually assess the degree of correlation between article characteristics. See
B.1. Determining Regression Models Between Article Characteristics
Process analysis system 100 also includes regression module 104 operative to determine the regression model between selected article characteristics. As discussed above, display module 108 is operative to generate a graphical display of regression models and present them on display device 818. See
The display of the relationship between two article characteristics in this manner provides useful information to process operators, design engineers and others associated with the design and manufacture of the article. The slope (steepness) of the regression line can be used to determine the relative sensitivity of article characteristics to changes in process settings. In addition, the slope (steepness) of the regression line can be used to identify article characteristics that will be more restrictive on (more sensitive to) the allowable range of process settings when specification limits are considered (see below).
B.1.a. Locating the Target Intersection
As discussed above, the design of an article generally results in a target value, as well as upper and lower specification limits, for each article characteristic (or at least the critical article characteristics). In one form, process analysis system 100 is operative to determine the intersection of the target values for a pair of article characteristics relative to the corresponding regression model.
In addition, as
In one embodiment, an Offset Table is created based on how far the regression line is located from (offset from) the target intersection for each article characteristic. The offset is presented in three formats: in the X-direction, the Y-direction and the direction perpendicular to the regression line.
B.1.b. Specification Limits
Process analysis system 100 is also configured to locate the upper and lower specification limits for the Y-axis article characteristic relative to the regression model between the Y-axis article characteristic and an X-axis article characteristic. See
In addition, process analysis system 100 is also operative to locate the upper and lower specification limits for the X-axis article characteristic relative to the regression model. This representation allows one to determine whether the regression line passes through the acceptable region bounded by the four specification limits. In other words, this representation allows for a determination as to whether it is even possible, given the current process and process inputs, to manufacture the parts within specification limits. In addition, locating the specification limits relative to the regression model allows for a determination of the maximum and minimum values (and, therefore, range) for the X-axis characteristic that will yield articles where the Y-axis characteristic is within specification limits. This range determination allows a manufacturer, for example, to determine if the part is in compliance with the specification limits for both the X- and Y-axis characteristics only by measuring the X-axis characteristic. To compute the minimum X-axis article characteristic, process analysis system 100 computes the value of the X-axis article characteristic at which the regression model intersects the lower specification limit for the Y-axis characteristic. Similarly, to compute the maximum X-axis article characteristic, process analysis system 100 computes the value of the X-axis characteristic at which the regression model intersects the upper specification limit for the Y-axis characteristic. In either case, the X-axis characteristic can be no larger than the upper specification limit for X and can also be no smaller than the lower specification limit for X.
B.1.c. Prediction Intervals
In addition, prediction intervals may also be used in the determination of minimum and maximum values for the X-axis characteristic (see Section II.B.1.b., supra). As
B.2. Predictor Characteristic
An embodiment of the present invention applies correlation and regression analysis to determine predictor characteristics in manufacturing processes. In one embodiment, a predictor characteristic is selected from the plurality of article characteristics associated with a part and used as the single X-axis characteristic. As discussed in more detail below, the predictor characteristic is chosen based on an assessment of the capability of a given article characteristic to be a predictor of other article characteristics. The selection of a predictor characteristic, therefore, reduces the number of article characteristic combinations that must be analyzed to a relatively small subset. For example, a part having 31 article characteristics would require analysis of over 900 relationships between article characteristics. The selection of a predictor characteristic reduces this to 30 combinations. In addition, the selection of a predictor characteristic can be used in a variety of ways to facilitate design, production, and measurement tasks associated with manufacturing. For example, a predictor characteristic can be used to greatly reduce the time and expense associated with measuring parts, as only the predictor characteristic needs to be measured during production to determine if all other article characteristics are within specification.
With a selected predictor characteristic, interface application module 110 directs regression module 104 to determine the regression model between the predictor characteristic (in one embodiment, as the x-axis characteristic) and all or a subset of the remaining article characteristics (see steps 206, 208, 210 and 211). When complete, the user is prompted to select a predicted article characteristic (step 212). Display module 108, in one embodiment, based on the equation defining the regression model, generates a graphical display of the regression model between the predictor characteristic and the selected predicted characteristic (step 214).
In addition to the regression model, display module 108 is further operative to add additional features to the graphical representation presented to users.
A variety of interface displays are possible. For example, the equation defining the regression model may be displayed to the user. Moreover, the maximum and minimum predictor characteristic values may be displayed to the user, as well as any other data associated with the article characteristics and/or the relationship between them. In one embodiment, the graphical user interface presented on display device 818 allows the user to select which of the above graphical elements to display.
B.2.a. Selecting Predictor Characteristic
The predictor characteristic can be selected using either a heuristic or a statistically-based approach. Moreover, the selection of a predictor characteristic may be based on a visual assessment of the correlations between article characteristics or an analytically-based assessment.
B.2.a.1. Graphical Selection
In one embodiment, a user can use the scatter diagrams to visually assess the degree of correlation, amounting to a visual estimation of the correlation coefficient for each scatter diagram. The closer the boundary or perimeter around the data points approaches a straight line, the higher the correlation coefficient. The exception to this general rule is for situations where the regression line is horizontal, or nearly so. See Section II.B., supra. The user can assess the scatter diagrams of all possible combinations of article characteristics. However, in another embodiment, the number of scatter diagrams used could be greatly reduced by picking one article characteristic to act as the foundational variable. Using the foundational variable as the X-axis variable, a scatter diagram would be then created for each remaining article characteristic, which would be plotted on the Y-axis. Picking the “foundational” (equivalent to the predictor) article characteristic can be based on looking at the “scatter” of the data, or can be randomly picked. While a visual assessment may be practical if a small number of article characteristics are involved, larger numbers of article characteristics, resulting in combinations into the thousands, requires (at least for practical purposes) the use of a computing device to analytically select the predictor characteristic.
B.2.a.2. Analytical Selection of Predictor Characteristic
To facilitate selection of a predictor characteristic, in one embodiment, correlation module 108 calculates the correlation coefficients between all or a subset of the article characteristics; determines, based on the calculated correlation coefficients, a value indicating the predictive capability of a first article characteristic relative to all other article characteristics; and repeats this process for all or a subset of the article characteristics.
Correlation module 106 then ranks the article characteristics according to the values computed in step 404 (step 406). Display module 108 then displays the ranked list on display device 818 to allow the user to select a predictor characteristic based at least in part on the predictive capabilities of the article characteristics (step 408). According to one embodiment, the user makes his choice (step 410), causing interface application module 110 to direct regression module 104 to compute the regression models between the selected predictor characteristic and the remaining article (predicted) characteristics (see above).
The correlation coefficient table may be populated using any suitable process or technique. However, in a preferred embodiment, correlation module 106 executes the methodology described below.
B.2.a.3. Population of Correlation Coefficient Table
According to standard industry practice, the data for a single article characteristic (e.g., one dimension) is vertically arranged in a column. Thus, each column stores the data for one and only one article characteristic. The resulting measurement data array will, therefore, have as many columns as there are article characteristics. More often than not, the data is stored in an Excel spreadsheet, or other suitable file format.
With this convention, then, each column represents a different article characteristic. Each row represents the article characteristic data (multiple article characteristics) for a single part. In the case of injection molding, each row stores the data associated with a single press operating cycle. If the mold is a single-cavity mold, each row will contain measurement data for a single part. However, if the mold is a 4-cavity mold, each row stores the measurement data for all four parts produced during one machine cycle. Typically, the same article characteristics are measured for each part in a multi-cavity mold; however, there is no constraint that requires this.
In one embodiment, correlation module 106 includes functionality to determine the correlation coefficients (according to standard statistical methods) among all article characteristics, compute a value indicative of the predictive capability of each article characteristic, and rank the article characteristics according to their relative predictive capabilities.
After correlation module 106 reaches the last remaining article characteristic (step 514), it fetches the computed correlation coefficients in the first column (A=1), transposes the column into a row, shifts the row by one relative to the correlation coefficient table and stores the data in the appropriate cells of the table (step 518). Correlation module 106 then increments the cell column position (A=2) and the article characteristic identifiers (X=2) (step 522), (Y=3) (step 508) and sets the cell row position equal to the column position (B=2) (step 524). Correlation module 106 then computes the correlation coefficient between the second article characteristic (X=2) and the third article characteristic (Y=3; step 508) (step 510) and stores it in the appropriate cell (A=2, B=2) (step 512). Correlation module 106 repeats this process until the correlation coefficient between the second-to-last article characteristic value and the last article characteristic value has been computed and stored (see step 520). As
As is apparent from the above-provided description, the process for populating the correlation coefficient table achieves a 50% reduction in the number of correlation coefficients that must be computed, because for every XY correlation, there is a corresponding YX correlation. It is also apparent that the compact notation of the correlation coefficient table greatly facilitates programming the sub-routine that populates the table. If the computations were to be done with only two correlations in each row (XY and YX), there would be over 600 rows for 50 part characteristics. The preceding method for populating the correlation coefficient table is one embodiment of the table population/compression algorithm. As previously mentioned, to maintain the usual and customary convention consistent with industry standards, the data for a single article characteristic is vertically arranged in a column. The methods described herein would function equally well if the data for a single article characteristic was horizontally arranged in a row and the algorithm was adapted for that data structure. For that case, the average correlation coefficient would be computed by taking the average of a row of correlation coefficients rather than a column.
B.2.a.4. Alternative Embodiment
In one embodiment, a user may use the functionality discussed above to complete the selection of a predictor characteristic and view the scatter diagrams with the predictor characteristic as the x-axis variable and including specification limits, and optionally prediction intervals. Based on such scatter diagrams, the user may select the predicted characteristics that are robust (insensitive to changes in process settings) and eliminate such article characteristics from the data set to eliminate “noise.” This selection can also be accomplished analytically based on the slope and Y-intercept of the regression model, the location and slope of the prediction intervals and the value of the upper and lower specification limits for both the x-axis and y-axis variables. In one form, such article characteristics have prediction intervals that do not intersect the predicted article characteristic specification limits. In this context, they constitute “noise” in selection of a predictor characteristic. The user then re-runs the selection of the best predictor based on the revised (reduced) data set.
B.2.a.5. Overriding The Statistically Best Predictor
In one embodiment of the present invention, process analysis system 100 allows the user to override the selection of the statistically best predictor characteristic.
B.2.b. Minimum and Maximum Values for Predictor Characteristic
To facilitate understanding of an embodiment of the present invention, it is useful to consider a simple situation where there are only two article characteristics of interest. One of the article characteristics is selected as the predictor characteristic using methods described above. The regression model establishes the relationship between the predictor characteristic and the remaining article or predicted characteristic. This situation is illustrated in
Because the regression model seldom has a perfect degree of correlation, there is uncertainty in using one part characteristic to predict the other. The prediction intervals associated with the regression model place limits on the uncertainty associated with predicting one part characteristic given a value for the other part characteristic.
B.2.c. Constraint Table for Predictor Characteristic
As discussed above, the foregoing discussion in section II.B.2.b., supra, involved a simplified situation involving only two article characteristics. In actual practice, a given part often has a large number of article characteristics of interest. In one embodiment, process analysis system 100 is further operative to create a Constraint Table. The Constraint Table contains, for each predicted article characteristic, the minimum (P-min) and maximum (P-max) values for the predictor characteristic as determined above (see Sections II.B.1.b. & c., II.B.2.b., supra).
From the Constraint Table (see
As described more fully below, process analysis system 100, in one embodiment, is operative, for each predicted article characteristic, to 1) determine whether the potential for defects exists; 2) determine whether the predicted article characteristic is robust; and, 3) if neither relationship 1) nor 2) is present, determine the maximum and minimum allowable values (Pmax and Pmin, respectively) for the predicted characteristic. In graphical and in various computational terms, process analysis system 100 locates, relative to the regression model (and the area it defines) between the predictor characteristic and the predicted article characteristic, a compliance area consisting of the area bounded by the upper and lower specification limits for both the predicted article characteristic and for the predictor characteristic (see
Process analysis system 100 then compares the bounded regression area to the compliance area to assess the relationship between the two areas. If the bounded regression area lies completely within the compliance region, the predicted article characteristic is robust. As to this relationship, process analysis system 100 sets the minimum and maximum predictor characteristic values associated with the first remaining article characteristic to the lower and upper specification limits, respectively, of the predictor characteristic. If any portion of the bounded regression area extends above and/or below the compliance area over the specification limit range of the predictor characteristic, a defect potential exists, causing process analysis system 100, in one embodiment, to report a defect condition.
Furthermore, one skilled in the art will recognize that 1) one of the three possible relationships (robust, defect, and constraining) discussed above must exist, 2) the three possible relationships discussed above are mutually exclusive, and 3) if two conditions are known not to exist, the third condition must exist. Accordingly, when one of the possible relationships is identified as existing, process analysis system 100 need not test for the existence of the other relationships. In addition, process analysis system 100 can test for these relationships in any desired order.
As one skilled in the art will recognize, the are many ways to test for the relationships discussed above. For example, process analysis system 100 reports a defect potential as to the predicted article characteristic, if all vertical cross-sections of the bounded regression area are fully or partially outside the compliance area. A predicted article characteristic can also be determined to be robust, if all vertical cross sections of the bounded regression area lie completely within the compliance area. Lastly, the predicted characteristic constrains the predictor characteristic when at least one vertical cross-section of the bounded regression area is fully inside the compliance area and at least one vertical cross-section of the bounded regression area is fully or partially outside the compliance area.
Process analysis system 100 can also determine whether the upper and lower prediction boundaries intersect the compliance area using known computational methods. For example, if the upper and lower boundaries of the bounded regression area lie within the compliance area, the predicted article characteristic is robust as to the predictor characteristic. If either or both of the upper and lower boundaries of the bounded regression area do not intersect the compliance area, then a defect potential exists. If the upper boundary of the bounded regression area intersects the upper boundary of the compliance area and the lower boundary of the bounded regression area lies within the compliance area, a constraining relationship exists. Lastly, if the lower boundary of the bounded regression area intersects the lower boundary of the compliance area and the upper boundary of the bounded regression area lies within the compliance area, a constraining relationship exists.
In another embodiment, process analysis system 100 can test to determine the values of the upper and lower prediction boundary values at the upper and lower specification limits of the predictor characteristic and determine whether these values (coordinates) lie within the compliance area. The following provides illustrative examples for didactic purposes.
If a defect potential exists, process analysis system 100 reports, in one embodiment, the defect potential by setting Pmin and Pmax to “DEFECT” as to the instant predicted characteristic (step 708). If no defect condition is detected, process analysis system 100 then determines, in one embodiment, whether the predicted characteristic is robust at least within the specification limit box 250 (step 710). In one embodiment, a separate method or function tests for such robustness by determining whether the boundary values discussed above lie within the specification limits of the predicted characteristic. In addition,
Otherwise, if no defect condition is detected and the predicted characteristic is not robust, process analysis system 100, in one embodiment, computes Pmin and Pmax (step 714). See Section II.B.2.b., supra. As
B.2.c.1. Restricted or Constrained Article Characteristic Data
In one embodiment, the present invention changes it's computational algorithms when the user supplies constrained or restricted dimensional data. Data can be constrained when the data range over which variation was induced is constrained by one or more factors. Examples of such factors might be personnel safety, harm to tooling, harm to process equipment, or manufacturing an incomplete part. For example, in an injection molding process, pressures, temperatures or speeds could be constrained to be below certain values. At pressures, temperatures or speeds above these constrained values, in the didactic example, plastic could be forced between mold halves (resulting in flash) and/or the mold could be damaged. In-the-other direction, pressures, temperatures or speeds could be constrained to be above certain values. At pressures, temperatures or speeds below these constrained values, insufficient plastic could be injected into the tooling resulting in a partially filled or incomplete part. These constraining factors result in dimensional data on the predictor dimension that have a maximum value that the user has decided will not be exceeded and/or a minimum value the user has decided the predictor dimension will not be less than.
In one embodiment of the present invention as illustrated in
B.2.d. Determining Producibility Targets and Ranges
Further parameters can be derived from P-min* and P-max* that will be useful for facilitating design of articles and process inputs and for setting process control variables. The maximum allowable range (P-range*) can be computed by subtracting the most constraining minimum value (P-min*) from the most constraining maximum value (P-max*).
In addition, a predictor production target (P-target*) can be determined. P-target* is the point that one would pick as the target for the average output of the process. It is, in one embodiment, the best “producibility” point that will maximize the chances of producing parts that are in compliance with specification limits. When properly selected, P-target* will minimize the percentage of data points that are outside of P-range* during production.
To avoid confusion between design target and a target for the predictor dimension (P-target*), a difference in terminology should be noted. The predictor characteristic (P) will almost always have an engineering design target (P-target). The engineering design target (or nominal value) is a value called for by a design engineer (e.g., a number on a drawing or in a specification). In contrast, P-target* is a target operating point for the process that will optimize production output, as discussed above.
In one embodiment. the predictor production target (P-target*) is selected as the midpoint of P-range*. See
B.2.e. Offset Table
As discussed in more detail below, process analysis system 100, in one embodiment, is further operative to generate an Offset Table indicating the amount by which a pre-process dimension would have to be adjusted to achieve a given design target. In other words, the Offset Table provides a value corresponding to the distance the regression line would have to be moved to pass through the target intersection. For example, in injection molding processes, an offset value is the amount by which a given mold dimension would have to be changed (for example, by adding steel through welding or removing steel through machining) so it would be possible to achieve the target intersection between the predictor dimension and a given predicted dimension). This is critical information for parts manufactured for the medical industry, for instance, where molds are sometimes modified to improve a part dimension by as little as 0.001 inch. This information can be used by the mold designer and the mold manufacturer. For other processes like plating, the Offset Table can tell engineers the required changes to the pre-plating dimensions of the part. For semi-conductor manufacturing, the Offset Table may inform the user the magnitude by which one or more mask dimensions must be adjusted. In addition, for processes involving CNC cutting or milling systems, the pre-process settings may be the dimensions programmed into the CNC application to create a desired article. The Offset Table can inform the engineers or CNC operators of the extent to which one or more coordinates or other settings require adjustment.
In one embodiment of the present invention, the magnitude and direction of the offsets in the Offset Table are computed by determining the vertical distance between the target intersection and the regression model. In a second embodiment of the present invention, the magnitude and direction of the offsets in the Offset Table are computed by determining the horizontal distance between the target intersection and the regression model. In a third embodiment of the present invention, the magnitude and direction of the offsets in the Offset Table are computed by determining the distance between the target intersection and the regression model in a direction perpendicular to the regression model.
The information contained in the Offset Table is also useful from yet another perspective. Whether the predicted characteristic value is above, at, or below the design target value for that predicted characteristic depends on the specific values selected for each of the process settings. In-other-words, the offset of the achieved predicted value from the design target value is dependent on the values selected for the process settings. For one set of process settings, the decision might be made to increase a mold dimension by removing steel. For a different set of process settings, the decision might be made to decrease that same mold dimension by adding steel. As can be imagined, this is an undesirable state of affairs. Using the new technology described herein, offsets are computed by determining the distance and direction of the regression model from the target intersection. Using the methodology described herein, offsets can be determined that are independent from the values selected for the process settings.
The Offset Table, in one embodiment, is particularly useful because it enables the tooling engineer to determine offsets independent of the particular process settings chosen by the operator or process engineer to generate parts. This is illustrated in
This contradictory situation in the prior art is explained in
B.2.f. Design Tolerance Relaxation Table
In many cases, it is faster and cheaper to relax the design tolerances on a given part instead of changing pre-process elements, such as the dimensions of a mold. The design engineer must, of course, make a decision as to whether or not this is feasible in light of the potential impact of tolerance relaxation for any given article characteristic on form, fit or function. Further, in some instances, the operating range (Prange*, supra) may be of such small size as to make it difficult, unlikely, or even impossible to produce parts or assemblies that have all article characteristics within specification limits. If the decision is made to increase producibility by increasing the operating range by relaxing tolerances, then one embodiment of the Design Tolerance Relaxation Table facilitates an assessment of which design tolerances to relax by providing a prioritized list of the optimal order in which tolerances should be relaxed, as well as an analysis of the increase in operating range achieved for each incremental tolerance relaxation.
In one embodiment, process analysis system 100 is operative to generate a Design Tolerance Relaxation Table which facilitates analysis of the achievable gains in operating range associated with relaxing the design tolerances of each predicted article characteristic. The most constraining predicted characteristic (i.e., having the highest Pmin or lowest Pmax) should be relaxed first. Said differently, the first tolerance to be relaxed is the tolerance on the article characteristic that most constrains the predictor characteristic. The Design Tolerance Relaxation Table then tells the engineer how much increase there would be in the operating range (the difference between P-min* and P-max*) as a result of relaxing that tolerance. The Design Tolerance Relaxation Table also tells the engineer the cumulative gain achieved by relaxing each variable in turn.
Process analysis system 100 then calculates the individual gains achieved by relaxing Pmax and, hence, the specification limits for the predicted article characteristics (step 836). In one embodiment, process analysis system 100 sorts the article characteristics by most constraining (lowest) Pmax value and then adds the predictor characteristic identifier and associated upper specification limit to the end of the sorted list. Process analysis system 100 then computes the individual gain achieved by relaxing the specification limit of a given article characteristic to the value corresponding to the Pmax of the next article characteristic in the sorted list. For example, the individual gain achieved by relaxing the specification limit of Var2 to the level of Pmax corresponding to Var4 is the difference between Pmax of Var2 and Pmax of Var4. This difference is 0.0006 inches; note that there is no computer round-off in this instance for the values of Pmax. The succeeding individual gains for the remaining article characteristics are computed in a similar manner until the last (predictor) characteristic is reached. Process analysis system 100 then computes the cumulative gains associated with each successive tolerance relaxation for both the Pmin and Pmax relaxations (step 838).
As one skilled in the art will recognize, the Design Tolerance Relaxation Table does not display the lower/upper specification limits for the predicted characteristics resulting from a relaxation of these limits to a given Pmin or Pmax; rather, the Design Tolerance Relaxation Table set forth in
In one embodiment, process analysis system 100 is operative to calculate the resulting lower and/or upper specification limits after the user has selected which article characteristic specification limits to relax. For example, the user may elect to relax only lower specification limits, only upper specification limits, or a combination of upper and lower specification limits to increase the allowable operating range (P-range*). In one embodiment, process analysis system 100 prompts the user for a selection of an article characteristic from the Pmin column and/or an article characteristic from the Pmax column of the Design Tolerance Relaxation Table and then calculates the new lower or upper specification limits for each article characteristic in the sorted list(s) up to the selected article characteristic(s). Specifically, for each article characteristic in the sorted list up to the selected article characteristic, process analysis system 100, using the regression model, calculates the article characteristic values where the boundaries (upper and lower prediction intervals) of the regression model intersect Pmin or Pmax as appropriate. For didactic purposes,
If the user selects a predicted article characteristic to relax Pmax (step 862), process analysis system 100 sets Pmax to the Pmax value corresponding to the next article characteristic in the sorted list (step 864). Process analysis system 100, for each article characteristic up to and including the selected article characteristic (see step 866), then computes the resulting new specification limits. Specifically, process analysis system 100 determines the slope of the regression model to select the lower or upper prediction interval, as appropriate (see above) (step 867) and computes the value at Pmax of the appropriate prediction interval (boundary) of the regression model (step 868) and stores the new specification limit in a table or other data structure (step 870) for ultimate display to the user. As one skilled in the art will recognize, use of the Design Tolerance Relaxation Table in this manner will result in new Pmin* and/or Pmax* values.
In another embodiment, process analysis system 100 displays the tolerance relaxations resulting for all predicted characteristics up to the predicted characteristic in a triangular array. The Constraint Table in
In one embodiment of the present invention,
As discussed above,
Process analysis system 100, in one embodiment, further accounts for the situation where the data is constrained (see Section B.2.C.1, above) and the computed tolerances may be less than the design tolerances. That is, if the data is constrained and the new upper tolerance is less than the original upper design tolerance (952), process analysis system sets the Upper Specification Limit, USL(I,J), to the original design upper specification limit, DesignUSL(I,J) (954) and the upper tolerance, UT(J), to the original design upper tolerance, DesignUT(J) (956).
C. Using the Maximum Allowable Range (P-range*) and the Predictor Production Target (P-target*)
For didactic purposes, it will be useful to clarify and define certain parameters as follows:
Knowing the maximum allowable range (P-range*) for the predictor characteristic is extremely useful. The actual process output will exhibit a certain amount of variability (VAR) and there will be a value that represents the predictor characteristic average process output (X-BAR). Having this information facilitates changing at least one process control setting in a manner that maximizes the likelihood that the process generates parts that are in compliance with the specification.
If it has been decided that P-range* is too “constrained”, then a second part characteristic can be measured to “open up” the “constraints” on P-range*.
There is also great utility in comparing the size of the actual process variability (VAR) for the predictor characteristic to the maximum allowable range (P-range*). The following situations are possible:
If the actual process variability (VAR) is smaller than and lies within the maximum allowable range (P-range*), then nearly all parts will be in compliance and there will be a greater margin of safety against errors or shifts in process output.
The present invention creates, for situation number 3. an excellent opportunity for the process engineer to investigate shifting the average process output (X-BAR) closer to the engineering design target (P-target) for the predictor part characteristic.
There is also great utility in comparing the average process output (X-BAR) for the predictor characteristic to its production target (P-target*). The following situations are possible, assuming that the process output distribution is symmetrical and the predictor characteristic target value (P-target*) is set at the midpoint of its maximum allowable range:
The closer the average process output (X-BAR) for the predictor characteristic is to the predictor production target (P-target*), the greater the likelihood that the process will produce parts that are in compliance.
When the average process output (X-BAR) is equal to the predictor production target (P-target*), the chances are maximized that the process will produce parts that are in compliance.
Similar conclusions can be reached even if the process output distribution is not symmetrical. In this situation, P-target* should be set at the point where the tails of the distribution have equal areas outside of P-range*.
Thus, the present invention facilitates determining the difference between the average process output (X-BAR) and the predictor production target (P-target*). This difference establishes both the magnitude and the direction that the average process output (X-BAR) should be shifted. With this knowledge, it is then possible to adjust one or more process control settings to move the average process output along the regression line to or closer to the predictor production target (P-target*).
The present invention provides further utility. It now becomes possible to determine whether the actual process variability (VAR) is too large relative to the maximum allowable range (P-range*). If this is the case, then a first option is to reduce the process variation. A second option is to increase the magnitude of the design tolerances. A third option is to do some combination of the previous two alternatives. The present invention can greatly facilitate efficiency and cost savings by requiring that the various process capability analyses discussed in this section be performed only one time for only the predictor characteristic instead of the 30 or 40, or however many total part characteristics, involved.
Moreover, the Constraint Table values provide other useful information for the design engineer. For example, if the decision is made to increase the size of the design tolerances, the Constraint Table facilitates a prioritized determination to be made as to 1.) which specification limit (upper or lower) for 2.) which article characteristic is most constraining and should be the first to be relaxed. This step can be repeated as often as desired, working “outward” from the most constraining to the least constraining part characteristic. The design engineer can also assess the impact of relaxing each tolerance on product performance and factor this information into the decision-making process.
The present invention creates still more utility. The design engineer now has information that enables a study to be conducted that evaluates the tradeoff between product performance and producibility. In addition, the design engineer may also, if circumstances permit, change the design target to the determined predictor characteristic target (P-target*), and make changes elsewhere in the system to compensate (if compensation is even required, for the change in the design target).
To condense the next series of comments, P-range*, VAR, and TOL (the difference between the upper and lower specification limits), will be represented by A, B, and C. In a similar fashion, P-target*, X-BAR and P-target will be represented by X, Y, and Z. The present invention facilitates the following comparisons:
As discussed above, in one embodiment, variation is induced in the part characteristics during an experimental production run. One of the by-products of that experimental production run is that the user learns which process settings have a major impact on the part characteristics. That knowledge enables the user to adjust a small number of process settings to position product output at any predetermined point along the regression model. In the case of injection molding, for example, the user may find that changing just one pressure setting, or one temperature setting, or one speed setting will be sufficient to translate the joint output of the article characteristics along the regression line.
Furthermore, the user may select a predictor characteristic (636) based on the correlation and regression analysis (612) to facilitate measurement of production output (630). For example, by analyzing the remaining article characteristics in relation to the predictor characteristic the user may identify robust predicted characteristics (i.e., article characteristics that will always be within tolerance limits) and eliminate them from measurement (step 634). Alternatively, or in conjunction therewith, the user may determine the maximum allowable range for the predictor characteristic and determine whether production output 630 complies with specification limits by measuring a single article characteristic (the predictor characteristic) (632).
As discussed above, to shift output along the regression line, an injection mold operator can change one or more process control variables, such as pressure, temperature, speed, etc. For illustrative purposes, an example of changing process inputs in the injection molding industry would be to change an internal dimension of a mold cavity. The regression line can be shifted vertically. In
An alternate method of shifting the regression line in
Yet another method of shifting the regression line is to create the shift by changing some combination of mold dimensions for both the predictor and at least one of the remaining article characteristics. In the specific example shown in
D.1 Simulation Mode
The previous description of process analysis system 100 described process analysis system 100 operating in an analytical mode. Process analysis system 100, in one embodiment, also operates in a simulation mode that simulates the operation of the manufacturing system used to produce the articles being analyzed and the measurement system used to measure the article characteristics, including the manufacturing process, the measurement process, tooling, pre-process dimensions such as molds, tooling, fixtures, software and CNC programming, material responses and natural variation.
In one embodiment, process analysis system 100 includes a process simulation module 200 that allows the user to simulate the effect of contemplated changes—(i.) to engineering design targets, (ii.) to engineering design tolerances, (iii.) to pre-process dimensions and (iv.) to the measurement system (i.) without changing design targets, (ii.) without changing design tolerances, (iii.) without modifying tooling, (iv.) without changing pre-process dimensions, (v.) without producing new parts, (vi.) without measuring article characteristics on new parts and (vii.) without changing the measurement system. The simulation module 200 of process analysis system 100 enables the user to verify whether or not the contemplated changes to design targets, design tolerances, pre-process dimensions and the measurement system will have the desired effect without incurring the time and expense involved in actually making the changes, producing parts, measuring part characteristics and then determining whether the changes accomplished the desired objectives.
As previously discussed, process analysis system 100 analyzes a data set of article characteristic values and allows the user to assess, among other things, the producibility of a given article at design target values and/or within design tolerances, as well as what changes could be made to increase the producibility of the article, such as changing pre-process dimensions and/or increasing the operating range through tolerance relaxations and/or bringing the output closer to target specifications and/or reducing variability. For example, as previously discussed, process analysis system 100 provides information such as Offset Tables detailing the amount by which a given article characteristic is offset from design target values. The tolerance relaxation tables allow the user to assess the effect of increasing tolerances on the operating range of the process. As discussed below, simulation module 200 of process analysis system 100, in one embodiment, allows the user to simulate changes to the design and to the manufacturing process, allowing the user to assess whether the contemplated changes achieved desired objectives (such as increasing the operating range or increasing quality by producing closer to design targets).
In one implementation, changes to the measurement system used to measure the article characteristic values can be implemented by increasing or decreasing the distance of the upper and lower prediction intervals from the regression model by some amount or percentage specified by the user. The user may then observe the changes to the operating range, the operating target, the Tolerance Relaxation tables and the like resulting from changing the magnitude of the prediction intervals. In one embodiment, the user can simulate increases in measurement system accuracy by decreasing the magnitude of the prediction intervals. Similarly, the user can simulate decreases in measurement system accuracy by increasing the magnitude of the prediction intervals.
In a second implementation, the user can change the magnitude of the prediction intervals even though there is no change to the accuracy of the measurement system. In this implementation, changes to the magnitude of the prediction intervals will include a greater or lesser percentage of data points. In-other-words, changing the size of the prediction intervals will change the percentage of the distribution encompassed within the prediction intervals. This is also referred to as changing the confidence interval.
As discussed above, these simulated changes to the design, the pre-process dimensions and the measurement system are used to create a simulated set of data representing what the manufacturing process output would be using the revised parameters. This simulated data set is then provided by simulation module 200 as input to process analysis system 100. Process analysis system 100 then analyzes the revised data set. The results from process analysis system 100 using the revised data set provide a prediction of the part characteristics that will be generated by the manufacturing process using revised tooling and/or pre-process dimensions. If the examples displayed in
The following provides an outline of the overall process flow associated with use of the process analysis system to at first analyze an original article characteristic data set, and then to run a simulation based on modifications to either tooling and/or pre-process dimensions and/or design targets and/or design tolerances and/or measurement systems.
1. Display the analytical offsets from the target values (
2. Display the Pmin Tolerance Relaxation Table (
3. Display the Pmax Tolerance Relaxation Table (
4. Display the scatter plots (
5. Display the original design targets and design tolerances (
6. The user determines the correction factor for each part characteristic which affects, for each part characteristic, the pre-process dimension (
7. The user determines changes to design targets.
8. The user determines changes to design tolerances.
9. The user determines changes to the accuracy of the measurement system.
10. Simulation module 200 of process analysis system 100 receives modification parameters, including the correction factors for pre-process dimensions, changes to design targets changes to design tolerances
11. Simulation module 200 simulates the operation of the manufacturing (and measurement) process by applying the correction factors to the original article characteristic data set and creating a new article characteristic data set.
12. Simulation module 200 creates a new data set. The new data set is the new data input for process analysis system 100. The new data set contains the simulated article characteristic data set, the new design targets, and the new design tolerances, which incorporate any changes the user made.
13. Process analysis system 100 analyzes the new data set.
14. Process analysis system 100 creates a new set of output data.
15. Steps 1.-13. are repeated until the user is satisfied with the results.
16. Once satisfied, the user implements the desired changes to (i.) design targets, (ii.) design tolerances, (iii.) pre-process dimensions, and/or (iv.) the measurement system.
D.2. Analytical Simulation
In the implementation described above, the simulation relative to pre-process parameters, such as pre-process dimensions of tooling, etc., is accomplished by modifying the data set of article characteristic values according to the modification parameters specified by a user, and running the modified data set through the process analysis system 100 described above. In other implementations, however, this simulation can be done analytically by changing the coefficients, or other aspects, of the equation(s) that describe the regression models between article characteristics. Figure XD illustrates a linear regression model, which can be represented by the equation
In one implementation, process analysis system 100 computes the slope (m) and the y-intercept (b) in order to render the correlation and regression diagrams on a display. Accordingly, changes to the y-intercept value (b) shifts the regression model vertically.
When the regression line, evaluated at the target value for the predictor characteristic, is above the target intersection, the offset is positive. When the regression line, evaluated at the target value for the predictor characteristic, is below the target intersection the offset is negative. The magnitude and direction (positive or negative sign) of the offset can be obtained from the Offset Table (see above).
Subtracting the vertical offset (VO) from the y-intercept (b) of the equation defining the regression line moves the regression line in a direction closer to the target intersection. If the vertical offset is positive, subtracting a positive number from the y-intercept decreases the value of the y-intercept (b). The regression line shifts downwards closer to the target intersection (TI). Similarly, the regression line shift downwards (closer to the target intersection) when the positive offset is subtracted from the data set of article characteristic values.
For didactic purposes, the language contained in this application refers to the use of techniques to “locate” or “position” or “determine the intersection” or “determine the range” or other terminology that might be used from a graphical perspective. Virtually all analytical techniques described herein document can be accomplished either graphically or analytically. It is to be appreciated that analytical techniques can be used when graphical techniques are described and graphical techniques can be used when analytical techniques are described. Indeed, a preferred embodiment of the present invention performs all computations, calculations, locations and determinations using analytical techniques. Graphical displays are created for the convenience and understanding of the user.
Also for didactic purposes, the language in this application refers to a regression line. It is noted above that the regression “line” need not be a straight line but may be curvilinear. It should also be noted that the regression model is frequently shown, for didactic purposes, as being a single line. It should be noted the regression model in a preferred embodiment of this invention includes use of prediction intervals.
It is to be noted, that for didactic purposes, the effect of changes in process control settings has been illustrated in terms of their effect on a single remaining (predicted) article characteristic. It is to be understood that the effect of changes in process control settings can be determined for more than two article characteristics. Similarly, it is to be understood that the effect of changes in process inputs has been illustrated in terms of their effect on a single regression model. It is to be understood that the effect of changes in process control settings can be determined for more than one regression model.
For didactic purposes, it has been assumed that the objective of introducing changes in either the process control settings and/or process inputs has been to move the joint operating position(s) and/or regression model(s) closer to one or more target intersections. It is to be further understood that these changes can be made to move the joint operating position(s) and/or regression model(s) to any desired position.
For didactic purposes, the upper and lower specification limits associated with the article characteristics have been described as being constant values, resulting in a rectangular compliance area. The present invention, however, can also be applied to circumstances where one or more of such specification limits vary based on one or more factors, resulting in a compliance area having a trapezoidal (for example) or other shape.
Finally, for didactic purposes, as noted immediately above, it has been assumed that changes have been introduced to optimize article characteristics relative to one or more criteria. It is to be understood that the algorithms, models and concepts set forth herein can be used to achieve the opposite effect. For example, it is possible to determine the required change in a joint operating point and/or regression model position needed in order to achieve a desired change in a process setting and/or process input. One objective of doing this could be to move a process control setting away from a dangerous or harmful setting. Another objective of doing this could be to match the production process and/or engineering design parameters to specific process inputs such as pre-configured raw material shapes.
Lastly, although the present invention has been described as operating in connection with injection molding processes, the present invention, as discussed above, has application to a variety of processes. For example, the present invention has application to sheet metal forming, plating and semiconductor manufacturing, as well as any other process where a material is added, removed, or changed in form or structure. The present invention has application to other non-manufacturing processes where characteristics of the output are related. Accordingly, the present invention has been described with reference to specific embodiments. Other embodiments of the present invention will be apparent to one of ordinary skill in the art. It is, therefore, intended that the scope of the invention not be limited to the embodiments described above.